Skip to content
Apart Research
Apart Research
Contact us
Sprint projectOct 26, 2026Waterloo Canada

Undermining Coalitions: The Impact of Social Manipulation in Multi-Agent Games

Zhuoran Lyu · Team Colluding AI

Submitted to AI Collusion Research Sprint. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: Undermining Coalitions: The Impact of Social Manipulation in Multi-Agent Games

Open in a new tab (opens in new tab)
Code (opens in new tab)
Share

This project investigates whether targeted social interventions can disrupt established coalitions among large language model (LLM) agents. Using a modified version of Liar's Bar, we evaluated four strategies (private disclosure, conditional threats, public exposure, and deceptive provocation) against a no-intervention control. DeepSeek V4.1 Flash and Llama 3.1 8B were tested across 300 games.

Control games demonstrated substantial favoritism toward coalition partners. Although certain interventions increased individual challenges against partners, only warnings produced a statistically significant effect on DeepSeek's challenges against all partner plays after multiple-comparison correction. No intervention significantly reduced overall coalition favoritism. Private interventions also increased internal discussions about the undercover agent.

These findings suggest that communication-based interventions may influence specific agent behaviors without reliably disrupting coalition-wide favoritism, highlighting the challenges of mitigating collusion after coalition formation.

Reviews

Judging this Sprint?

Review “Undermining Coalitions: The Impact of Social Manipulation in Multi-Agent Games”

Your public critique appears on the project page without your name. Reviews of strong projects are shared with funders, as you choose below.

Not shown on the project page.

Shown on the project page, without your name.

Only the Apart team reads this, and funders if you agree below.

Decline to score

Only when the quality of the writing makes the work impossible to follow. Say why in your private critique; the project page shows only that a judge declined, not why.

How much would this matter for the field if it worked? How innovative is it? For scores of 4-5: is this actually new to the field, or replicating recent work?

Scoring guide
  1. 1Negligible. No clear problem addressed, or no meaningful novelty.
  2. 2Limited. Addresses a real problem but with a generic or well-trodden approach. Incremental at best.
  3. 3Moderate. Clear problem with a reasonable approach; some novelty in framing or method beyond routine application of existing tools.
  4. 4Significant. Important problem with an original approach, or identifies a neglected problem area. A valuable contribution others could build on.
  5. 5Exceptional. Tackles a critical problem with a genuinely novel approach, or opens a new research direction. Clear theory of change. You'd be excited to share this with researchers in the area.

How sound are methodology, implementation, and findings?

Scoring guide
  1. 1Seriously flawed. Methodology broken, results uninterpretable, or implementation doesn't work.
  2. 2Weak. Approach has significant gaps: missing validation, flawed experimental design, or incomplete implementation.
  3. 3Competent. Technically solid given the short duration. Methodology makes sense, results are interpretable, limitations acknowledged, work builds toward clear conclusions.
  4. 4Strong. Thorough methodology with convincing validation. Results clearly support conclusions. Immediately useful for future work.
  5. 5Exceptional. Ambitious scope executed rigorously. Surprising findings, novel methods, or unusually robust validation.

How clearly are work, findings, and impact potential communicated?

Scoring guide
  1. 1Incomprehensible. Cannot determine what the project is actually claiming or doing.
  2. 2Hard to follow. Key information buried, missing, or diluted by excessive length. Significant effort to extract main points.
  3. 3Clear enough. Can understand the problem, approach, and results without undue effort. Core content clearly present: problem, method, findings, limitations.
  4. 4Well presented. Easy to follow, well-structured, appropriate level of detail. Target audience would get it quickly.
  5. 5Exceptionally clear. A pleasure to read. Complex ideas made accessible. Could serve as a model for how to present this type of work.
Sharing your name *

When we share a strong project with funders, your name, scores and public critique go with it, confidentially; this choice is only about whether your name is public. If you think this project deserves funding, please also endorse it: that carries the most weight.

Share my private critique with funders *

See our Privacy Policy.

  1. This paper asks a fresh question for track 3.2 - once agents are colluding, can an undercover agent break it up? It builds directly on Liar's Bar and tests 4 social tactics - inform, warn, expose and provoke against a no intervention control. The design is the strongest I have seen so far - 300 seed-paired games across 2 models, sign-flip permutation tests with Holm correction, and a positive control showing allies really do favor each other (DeepSeek left 96.6% of partner lies unchallenged vs 77.7% for others). The paper also checked AI classified chat against human labels and all game logs are in the repo. The main takeaway is useful - an agent can challenge its ally more often with no drop in the coalition's overall favoritism.

    The main limitation is that each intervention changed information, threat. publicity and deception simultaneously so it is hard to tell what drove any effect. Also, coalitions were assigned by instruction instead of being formed by the agents themselves. The prompts had 2 inconsistencies as we ll - the random revolver and allies told nobody knew. DeepSeek’s control rate rests on only four challenges, and only one effect survived correction. The paper also designs deceptive manipulation tactics but has no dual-use discussion.

    I would suggest separating the parts of each tactic. For example the same warning with and without a real penalty, or the same message sent privately vs publicly. I would also recommend testing on coalitions where agents formed allies on their own as it will show if behavior changes outside of assigned alliance.

    Read full reviewShow less

Cite this project

@misc{lyu2026undermining,
  title = {{Undermining Coalitions: The Impact of Social Manipulation in Multi-Agent Games}},
  author = {Zhuoran Lyu},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/undermining-coalitions-the-impact-of-social-manipulation-in-multiagent-games-bhfa}},
  url = {https://apartresearch.com/sprints/projects/undermining-coalitions-the-impact-of-social-manipulation-in-multiagent-games-bhfa}
}

More from AI Collusion Research Sprint

All 392 entriesBrowse all projects

A Sprint is running now

AI x Epistemics Research Sprint · Nov 13 - 15, 2026